Inference
EN
Vals AI

Tax Agent Bench

Evaluating agents on research-grade US tax questions using the Tax Agent Bench harness

ModelsModels with a result.82
Top resultBest result on this benchmark.51.8%Gemini 4 Carbon
Top-3 spreadPoints from first to third.2.1 pts
UpdatedDate the source changed the results.9 Nov 2026

Result and cost

81 of 82 results

Results

82 results
#ModelResultThe score from the source.CostUS dollars to run one task.TimeTime to run one task.
01Gemini 4 Carbon51.8% ±3.6$6.3435 min
02Claude Fable 5.150.7% ±3.6$13.1829 min
03Qwen3.8 Max Internal49.7% ±3.6$4.0787 min
04Gemini 4 Argon49.2% ±3.6$3.4716 min
05Claude Opus 547.1% ±3.6$5.1417 min
06Claude Opus 5.544.5% ±3.6$14.4067 min
07GLM-5.343.3% ±3.6$1.8051 min
08Muse Wonder Full43.0% ±3.6$2.1518 min
09Muse Spark 1.3 Max42.4% ±3.5$0.394 min
10Claude Sonnet 5.542.0% ±3.6$8.7359 min
11Inkling 2 Small38.9% ±3.3$0.1447 min
12Claude Fable 538.3% ±3.5$6.7014 min
13Muse Ember38.2% ±3.5$0.0322 min
14Muse Spark 1.337.9% ±3.5$0.328 min
15Muse Wonder36.3% ±3.5$1.888 min
16Grok 4.735.8% ±3.5$1.8017 min
17Hy434.4% ±3.3$1.6963 min
18Grok 4.633.7% ±3.4$0.97811 min
19Kimi K333.4% ±3.4$2.7940 min
20Muse Spark 1.232.8% ±2.8$0.2692 min
21Hy4 preview32.6% ±3.4$0.58541 min
22MiMo-V2.6-Pro32.4% ±3.4$0.13426 min
23MiniMax M3.1 Flash32.1% ±3.3$0.32133 min
23Gemini 3.8 Flash32.1% ±3.4$0.8573 min
23Qwen3.8 Max32.1% ±3.4$1.5541 min
23Claude Fable 5.232.1% ±3.4$3.315 min
27Claude Haiku 5.531.9% ±3.4$1.1650 min
28GLM-5.3-Flash31.6% ±3.4$0.13822 min
29GPT-5.6 Sol31.1% ±3.3$7.5145 min
29Ember-131.1% ±3.3$2.9731 min
31Qwen3.8-27B30.9% ±3.3$0.81522 min
32Mistral Large 4 OSS30.7% ±3.3$0.50317 min
33Mistral Large 430.4% ±3.3$1.3957 min
34GPT-5.6 Terra30.0% ±3.3$4.8361 min
34Claude Opus 4.830.0% ±3.3$1.698 min
36Claude Sonnet 529.4% ±3.2$1.7919 min
37MiMo-V2.6-Flash28.5% ±3.1$0.04917 min
38DeepSeek V4 Flash 073128.1% ±3.2$0.1566 min
39Step 5 Preview27.3% ±3.1$0.55321 min
40DeepSeek V4 Pro 081326.8% ±3.1$0.72927 min
41DeepSeek V4.1 Flash 101726.6% ±3.1$0.1356 min
42Muse Spark Open26.3% ±2.7$0.88412 min
43Claude Sonnet 4.625.8% ±3.2$1.1912 min
44DeepSeek V4 Pro 042325.2% ±3.1$0.71921 min
45Grok 4.524.9% ±3.1$0.4745 min
46GPT-5.6 Luna24.8% ±3.1$0.46543 min
47Claude Opus 4.723.7% ±3.1$1.275 min
48Gemini 3.7 Flash22.8% ±3.0$0.48487 s
49GPT-5.521.8% ±3.0$4.0716 min
49GPT-6.1 Sol21.8% ±3.0$1.3220 min
51Gemini 3.5 Flash21.7% ±3.0$0.864 min
52GPT-6 Astra20.7% ±2.9$5.8015 min
53Ling 3.0 Flash Af Rc320.6% ±2.9$0.0256 min
53MiniMax M320.6% ±2.9$0.1645 min
55GPT-6 Luna19.9% ±2.9$0.19133 min
56Mercury 318.2% ±2.5$0.0464 min
57Gemini 3.6 Flash18.0% ±2.8$0.3562 min
58GPT-6.1 Terra17.1% ±2.7$2.1646 min
59Qwen3.7 Max15.8% ±2.6$0.4353 min
60Inkling-Small15.4% ±2.4$0.0926 min
61Inkling15.3% ±2.4$0.35414 min
62GPT-6 Sol15.0% ±2.6$2.2818 min
63Kimi K2.612.4% ±2.4$0.56615 min
64Gemini 3 Flash Preview11.8% ±2.3$0.1012 min
65Nemotron 3.5 Super VL 120B A12B11.6% ±2.2$0.04673 s
66MiMo-V2.5-Pro11.5% ±2.0$0.046 min
67Qwen3.7 Plus10.5% ±2.0$0.0945 min
68Gemini 3.1 Pro Preview9.3% ±2.1$0.3022 min
69Beam 1.08.8% ±1.9—58 s
69GPT-5.4 mini8.8% ±2.0$0.80617 min
71MiMo-V2.57.8% ±1.8$0.0178 min
72Qwen3.6 Plus7.7% ±1.8$0.2986 min
73Gemini 3.5 Flash-Lite7.3% ±1.9$0.08247 s
74Grok 4.20 (Reasoning)7.0% ±1.8$0.11989 s
75MiniMax M2.76.9% ±1.7$0.125 min
76Grok 4.36.2% ±1.7$0.1172 min
77Claude Haiku 4.5 Thinking5.9% ±1.5$0.2263 min
78GPT-5.4 nano5.0% ±1.4$0.074 min
79Mistral Medium 3.54.5% ±1.5$1.447 min
80Gemini 3.1 Flash Lite Preview3.1% ±1.3$0.01419 s
81Nemotron Lightning 3p5 30B A3B2.6% ±1.1$0.03120 min
82Mercury 2.52.5% ±1.0$0.02430 s